Decision Graph Protocol FAQ

September 19, 2026 · View on GitHub

What is DGP?

Decision Graph Protocol is an MIT-licensed protocol by Numerous ApS for decision-based AI agents. An application offers immutable evidence frames, typed decisions, and guarded transitions. Assessments describe judgments; commits request effects.

How does DGP relate to TypeSafe and Jev?

Jev is the intended first intelligent assessor. DGP remains assessor-neutral and can also use deterministic, human, or other model judgments. See the TypeSafe/Jev integration guide.

Does DGP replace an LLM agent or existing tools?

DGP supplies a decision/control contract around application capabilities. Existing tools and native interfaces can share the same host guards. An LLM can still prepare options, generate bounded text, or supply missing inputs.

Is DGP a workflow engine or a complete decision tree?

No. A publisher may expose only the useful current decisions in part of an application. Graph previews describe possibilities; they do not grant authority to execute future actions.

Does a high-confidence Jev answer authorize an action?

No. The application rechecks permissions, state, expiry, and domain guards. Confidence is an assessment property, not an execution credential.

Can DGP batch decisions or speculate?

The optional assessment-batching profile describes independent assessments with shared evidence and per-item outcomes. Speculative assessments cannot be directly committed. This profile has schemas and contract tests, not a production endpoint.

What can I run today?

ThreadDesk's local HTTP demo, browser UI, controller CLI, deterministic mock resolvers, and validation suite. Its Git/test/publication effects are simulated. Farm Energy is a design profile. Read implementation status.

Is DGP an official TypeSafe, MCP, or GraphQL standard?

No. It is an independent experimental Numerous ApS protocol, with documented interoperability references and optional mappings.